Information Networks: State of the Art Michael R. Berthold and Tobias Kötter The subject of the presentation is Information Networks: State of the Art

Information Networks: State of the Art Outline Information Networks Properties of Information Networks Information Unit properties Relation properties Prominent Types of Information Networks Ontologies Semantic Networks Topic Maps Bayesian Networks Bisociative Information Networks (BisoNets) Comparative Matrix Let me first give you a brief overview. I will start with a short introduction on information networks followed by properties of information units and relations that can be used to classify information networks. After that I will talk about prominent types of information networks such as Ontologies and the properties they support. Finally I will conclude with a comparative matrix of the supported properties per information network. Information Networks: State of the Art

Information Networks: State of the Art Composed of: Information Units Physical items, concepts, ideas, … Represented by vertices Relations Connections between Information Units Usually represented by edges Commonly used for data integration Well defined structure allows to discover pattern of interest extract network summarization visually explore underlying relations Information networks are composed of Information units and relations. Information units represent concepts, ideas, physical items and are represented by vertices. Relations represent connections between Information units and are usually represented by edges Information networks are mostly used for data integration due to their flexible structure. This flexible but well defined structure also allows to discover pattern of interest such as cliques or hubs, to summarize the network in order to get a better overview of the global structure and to explore the underlying relations. Information Networks: State of the Art

Properties of Information Units Named the name of the information unit Attributed E.g. link to original data or translations of the original label Might be considered while reasoning or analyzing the network Do not carry general semantic information Typed Allows to distinguish between different semantics of information units Can additionally be organized in a hierarchy or ontology Hierarchical Subgraph Represents more complex concepts Coming next to the properties of information networks starting with the properties of information units. The first property allows to attache attributes to information units such as links to original data. This information can be considered … but do not carry general semantic information The second kind of property are typed information units allow to distinguish between… Hierarchical information units are subgraphs that represent more complex concepts such as cellular processes Information Networks: State of the Art

Properties of Relations Attributed Can be considered during the reasoning process Do not carry a general semantic information Typed Distinguishes between different semantics of relations Can be organized in a hierarchy or ontology Weighted Measure of reliability Allows the integration of facts and pieces of evidence Directed Explicitly models relationships that are only valid in one direction Multi relation Multi edges supporting any number of members The second type of network properties that describe the relations. Attributes allow to attach additional information to a relation. They can be considered … but do not… Typed relations can be used to distinguish … Relation types as well as information unit types can be organized in a … A relation weight represents the reliability of a connection and allows to integrate not only facts but also pieces of evidence Directed relations model explicit relations that are only valid in one direction such as a parent child relationship. Multi relations support more than two members which can be necessary to describe gene expression experiments or authors of a paper Information Networks: State of the Art

Properties of Ontologies Relations Attributed Typed Weighted Directed Multi relation Information Units Named Hierarchical On the left you can see the properties of the information units and on the top the properties of relations. Information Networks: State of the Art

Information Networks: State of the Art Ontology Controlled vocabulary for information units and relations Requires comprehensive domain knowledge Mostly manual or semi-automatic created The first type of the prominent information networks I want to talk about are ontologies. They support typed and directed relations. They use a controlled… The creation of the vocabulary requires a comprehensive… Most of the Ontologies are manual… The example shows information networks and their possible properties Information Networks: State of the Art

Properties of Semantic Networks Relations Attributed Typed Weighted Directed Multi relation Information Units Named Hierarchical Information Networks: State of the Art

Information Networks: State of the Art Semantic Networks Types might be organized in an ontology URI used to identify information units and relations Usually based on Semantic Web technologies Resource Description Framework (RDF) Knowledge representation and storage framework Triples consists of subject, predicate and object RDF Vocabulary Description Language (RDF Schema) Defines a vocabulary to describe properties and classes Used to describe the members of a triple Web Ontology Language (OWL) Extends RDF Schema The second type of information network are the semantic networks. They support attributed information units as well as typed and directed relations The relation types are mostly organized in an ontology to be more expressive URIs are used to identify… Most of the semantic web systems base on semantic web technologies such as RDF for data storage RDF Schema or OWL to describe the used attributes Information Networks: State of the Art

Semantic Network: Example The example shows a parent child relationship between one parent with the name Parent an its two children child1 child2. The typical rdf triple can be seen with subject, predicate and object. Information Networks: State of the Art

Properties of Topic Maps Relations Attributed Typed Weighted Directed Multi relation Information Units Named Hierarchical Information Networks: State of the Art

Information Networks: State of the Art Topic Map Topic represents generally everything, a concept, an idea, … Topics have zero or more types assigned represented by topics Associations model relations between any number of topics Association have a type assigned represented by topics Association members play a certain role represented as topic Occurrences link topics with resources they stem from Occurrences have any number of types represented by topics Virtually everything in topic maps is a topic Information Networks: State of the Art

Information Networks: State of the Art Topic Map: Example In this example all topics are depicted as ellipses and as you can see basically everything is a topic. Topic, relation and occurrence types are depicted as dashed lines whereas the occurrence itself is depicted as dotted line. The example shows three genes that were overexpressed in the same experiment with name exp_1. Information Networks: State of the Art

Properties of Bayesian Networks Relations Attributed Typed Weighted Directed Multi relation Information Units Named Hierarchical Information Networks: State of the Art

Information Networks: State of the Art Bayesian Networks Vertices represents variables Relations and their direction model dependencies Relation weights represent probabilities Remarks: Bayesian Networks do not have weighted edges. The probabilities are attached to the nodes!!! The example depicts a typical Bayesian network with the probabilities of smoking, bronchitis, lung cancer and dyspnea. Information Networks: State of the Art

Properties of BisoNets Relations Attributed Typed Weighted Directed Multi relation Information Units Named Hierarchical The last network type I just want to mention briefly are BisoNets that have been developed by the Bison EU consortium. The main goal of these networks is to support the integration not only of facts but also pieces of evidence from heterogeneous data sources in order to find connections across diverse domains so called bisociations. Information Networks: State of the Art

BisoNets: Bisociative Information Networks k-partite graph Partitions represent types e.g. gene, document, … Nodes represent concepts, relations or BisoNets Edge weight represents the certainty of a connection Nodes might carry any number of attributes Remark: What is the difference between bisonets and relational data model? Is it necessary that partitions are aligned next to each other? Nodes represent concepts such in the previously mentioned network types But represent also relations which allowing the modeling of multi relations And can also represent BisoNets themselfe in order to represent a hierarchical structure. Information Networks: State of the Art

Information Networks: State of the Art Comparative Matrix Information Units Relations Attributed Typed Hierarchical Weighted Directed Multi relation Ontology Semantic Networks Topic Map Bayesian Networks BisoNets I want to conclude my talk with the comparative matrix of all presented network types. As you can see by the supported propeties ontologies, semantic web and topic maps were created to support the integration of semantic meaningful data sources such as facts. And Bayesian networks on the otherhand support the integration of probabilisitic data representing pieces of evidence. BisoNets combine all these properties in order to integration various kinds of data in order to find domain crossing connections. Thank you for your attention. Any questions? Information Networks: State of the Art

Thank you for your attention! Any questions? Information Networks: State of the Art